The Factory Data Problem ASEAN Manufacturers Do Not Talk About

The most common barrier to AI deployment in ASEAN manufacturing is not a lack of budget, executive will, or vendor availability. It is the state of factory data infrastructure. ASEAN manufacturer factory data infrastructure problems, siloed OT and IT systems, inconsistent sensor logging, disconnected ERP and MES layers, and absent data ownership frameworks, prevent AI from accessing the reliable inputs it needs to function in production. This post examines each problem specifically and explains how manufacturers can address them before investing in AI systems.

Why Factory Data Is the Starting Point

AI systems in manufacturing, predictive maintenance models, quality inspection algorithms, production optimisation tools, all have one structural requirement: consistent, labelled, historically continuous data. When that data does not exist in a usable form, an AI system cannot be trained, validated, or monitored in production.

Many ASEAN manufacturers have completed pilot AI projects that produced promising results in controlled conditions. The pilots then fail to scale because the production data environment does not match what was available during the pilot. This is the factory data problem: it is not visible until it blocks progress. This pattern appears consistently in manufacturing AI deployments across ASEAN, including programs structured around Industry 4.0 in Vietnam national roadmaps.

The Four Specific Problems

1. Siloed OT and IT Systems

Operational technology (OT) systems, PLCs, SCADA, DCS, and information technology (IT) systems, ERP, MES, WMS, are frequently separated by both infrastructure and organizational boundaries. Data generated on the factory floor does not flow automatically into the systems where business logic and historical records are stored.

In many ASEAN factories, this separation reflects both legacy procurement decisions and deliberate security choices. The result is that AI systems designed to connect operational performance with business outcomes have no reliable pathway to receive the inputs they need from both layers simultaneously.

2. Inconsistent Sensor Logging

Even where sensors are installed, data logging is frequently inconsistent. Common issues include variable sampling rates that change without documentation, sensors that go offline during shift changes and log no data rather than a null value, gaps created by network interruptions that are not recorded in any audit log, and sensor configurations that change when equipment is maintained, without version control of the configuration change.

These inconsistencies produce training data that contains hidden gaps and structural shifts. An AI model trained on this data may appear to perform well in development but produce unreliable outputs in production because it has learned patterns from data that does not accurately represent the underlying process.

3. ERP and MES Disconnects

Manufacturing Execution Systems (MES) track what happens on the factory floor: work orders, production quantities, quality holds, machine downtime. Enterprise Resource Planning systems (ERP) track the business layer: inventory, procurement, finance, customer orders.

In many ASEAN manufacturing operations, MES and ERP systems were purchased at different times, from different vendors, with custom integration work that has not been maintained. The result is that data from both systems cannot be reliably joined. A manufacturer who wants to train an AI system to correlate production parameters with cost outcomes faces a foundational data alignment problem before any model work can begin.

4. Absent Data Ownership

In many factories, no one owns the data. Engineering teams own the equipment. IT teams own the networks. Operations teams own the processes. Finance teams own the reporting. When an AI project starts, the question of who is responsible for data quality, data access, and data governance frequently has no clear answer.

This is not a technology problem. It is an organizational design problem that manifests as a technology constraint. Without assigned data ownership, issues with data quality go unaddressed, access requests stall, and the accountability structure needed to sustain an AI system in production does not exist.

How These Problems Connect

These four problems are not independent. Siloed OT and IT systems make consistent sensor logging harder to enforce. Inconsistent sensor logs make ERP and MES reconciliation unreliable. Absent data ownership means no one fixes the gaps when they are identified. The result is a compounding data infrastructure deficit that grows as the factory generates more data.

Manufacturers who try to address these problems only when an AI project is already underway face significant delays. The infrastructure work required takes time, typically several months to achieve baseline data reliability, and cannot be compressed by adding more technology vendors.

Where to Start

A practical starting point is a data infrastructure audit conducted before any AI vendor is selected. The audit should map all data sources by type, owner, sampling rate, and storage location; identify where OT and IT data can be joined and where it cannot; document all known gaps in historical sensor logs; and assign provisional data ownership to specific roles.

This audit does not require AI expertise. It requires engineering and operations knowledge combined with an honest assessment of the current state. Manufacturers who complete this audit before engaging AI vendors reduce the risk of expensive scope changes after a project has started.

Gradion works with ASEAN manufacturers on data infrastructure readiness before AI deployment. Our engineering teams have mapped OT and IT architectures in industrial environments across the region and understand the specific constraints that appear in facilities that were not designed with data integration in mind. Across manufacturing AI engagements in Vietnam and the broader ASEAN region, the most consistent finding is that data readiness determines project timelines more than model selection or compute resources.

Frequently Asked Questions

What is the factory data problem that prevents ASEAN manufacturers from deploying AI?

The factory data problem refers to a cluster of infrastructure challenges that prevent AI systems from receiving the reliable, consistent data they need to function in production. The core issues are siloed OT and IT systems, inconsistent sensor logging, ERP and MES disconnects, and absent data ownership frameworks. These problems are common in ASEAN manufacturing because factories were built over decades with systems purchased independently, and data integration was rarely designed in from the start.

How long does it take to fix factory data infrastructure before deploying AI?

The timeline depends on the severity of existing gaps and the complexity of the factory environment. Straightforward environments with moderate integration gaps typically require three to six months to reach baseline data reliability. Complex multi-site operations with legacy OT systems and fragmented ERP implementations may require longer. A data infrastructure audit conducted at the start of the process provides a more accurate estimate for a specific facility.

Can manufacturers deploy AI in one part of the factory while fixing data infrastructure in others?

Yes, in some cases. Predictive maintenance models for specific equipment categories can be deployed using sensor data from those assets alone, without requiring full OT and IT integration. However, AI systems that require cross-functional data, correlating production parameters with cost, quality, and procurement data, cannot be deployed reliably until the underlying integration gaps are resolved.

What is the difference between OT and IT in a manufacturing context?

Operational Technology (OT) refers to hardware and software that monitors and controls physical equipment: programmable logic controllers (PLCs), distributed control systems (DCS), and SCADA platforms. Information Technology (IT) refers to the business systems that manage data and processes at the organizational level: ERP, MES, WMS, and CRM platforms. In most factories, these two layers evolved separately and require deliberate integration work to share data reliably.

Who should own data governance in a manufacturing organization?

Data governance in manufacturing works best when ownership is assigned at two levels. At the asset level, engineering or operations teams should own sensor configuration, calibration records, and data logging standards for specific equipment. At the enterprise level, an IT or digital function should own data access policies, integration architecture, and data quality monitoring. Without both levels of ownership, data quality issues are identified but not resolved, because responsibility is unclear.